NEW YORK – With $460 million in gross proceeds from an initial public offering in July, Cambridge, Massachusetts-based Relay Therapeutics is beginning to test highly selective precision oncology drugs in human trials. The drugs were discovered with the help of a machine learning approach that models the motion of proteins.
Relay, founded four years ago, considers itself to be part of a "new breed of biotechs" that heavily lean on computer science, machine learning, and visual simulations to discover differentiated, novel anti-cancer drugs. The company's drug development platform, dubbed Dynamo, relies on the computational power of a custom-designed supercomputer, which can screen billions of compounds from a digital library and identify those which, on the basis of their binding activity to specific proteins detailed in visual, moving simulations, have the potential to treat cancer.
Using this platform, the company has already advanced two drugs into clinical development: the SHP2 inhibitor RLY-1971 and the FGFR2 inhibitor RLY-4008. The company has a third agent, a PI3Kα inhibitor called RLY-PI3K1047, in preclinical development, as well as three oncology programs that are not yet publicly disclosed.
In 2016, Relay Therapeutics launched with $57 million in series A funding led by venture capital firm Third Rock Ventures. One of Relay's four founders, D.E. Shaw & Co.'s David Shaw, gained recognition in the 90s as one of the pioneers of algorithmic trading on the stock market.
On the surface, that expertise may seem like a far cry from the company's goal of developing selective drugs for molecularly-defined subsets of cancer patients. But the novel algorithms and machine learning approach that once transformed Wall Street are, in many ways, the same concepts underlying the highly technical drug development approaches being explored at biotechs like Relay.
In addition to Shaw, several of the company's other cofounders and current executives have a computer science or computer-aided drug design background. According to the company's president and CEO Sanjiv Patel, of the 130 scientists now employed at Relay, a third are on the "computational side" and the remaining two-thirds work on the experimental side, conducting research to test and validate compounds as potential agents.
Together, the experimental and computational scientists are leveraging the supercomputer "Anton 2," designed by Shaw's research firm D.E. Shaw Research, to create visual simulations that provide insight into both the structure of full-length proteins and the ways in which the proteins move in the body over time and interact with other molecules. The targeted agents that the company develops are rooted in this visualization of protein motion.
The approach, which the company has coined Motion-Based Drug Design, is part of the Dynamo platform. "Dynamo" is what Relay calls the whole computer-enabled process of displaying full-length proteins, visually representing their motion through computer simulations, screening billions of compounds from a computer library, selecting the compound predicted to target a specific cancer-associated protein of interest, and visualizing the way the molecule and protein interact.
Using Dynamo, Patel said, Relay aims to "tackle the basic limitations of traditional structure-based drug design." Typically, in drug development, scientists tend to visualize protein structures in a more or less static way, which in his view is a major limitation. Visualizing proteins' patterns of motion over time can help scientists overcome these limits, he said, though the technology enabling such simulations only became available recently.
"You need a huge amount of computational power to create virtual systems of these proteins so that you can virtually test potential small molecules and how they would bind and interact [with these proteins] rather than just be reliant on the wet lab," Patel explained. That computational power, he said, became accessible to Relay via Shaw's custom supercomputers, which are designed to detail protein motion through the molecular dynamics process — that is, by calculating the force that each of a protein's atoms exerts. Once the computational power became available — and the potential to use it for precision drug development was realized — Relay was born. The company was listed on the Nasdaq in mid-July, its stock price rising from $20 to $35 by the close of its IPO.
Differentiating through selectivity
Although its approach to drug discovery and development may be novel, Relay is a player in precision oncology therapeutics, a market crowded with biotech startups developing drugs for indications defined by rare biomarkers. Certainly, the targets Relay has chosen to focus on — SHP2, FGFR2, and PI3Kα — are also of interest to many other drug developers.
The US Food and Drug Administration, for instance, has already approved the targeted agent erdafitinib (Janssen's Balversa) for patients with bladder cancers with FGFR alterations, and more recently, pemigatinib (Incyte's Pemazyre) for patients with bile duct cancers with FGFR fusions or rearrangements. But according to Relay's chief medical officer Don Bergstrom, Relay's FGFR2-inhibiting candidate RLY-4008 is unique in that it exclusively targets FGFR2 without binding to other members of the FGFR family, such as FGFR1 and FGFR3, which can cause significant toxicities for patients.
"This really solid target, FGFR2, hasn't been able to be fully leveraged as a precision oncology target, because no one's been able to truly selectively target it," Bergstrom said, explaining that the side effects of simultaneously targeting FGFR1, a protein which plays an important role in normal kidney function, limits the drug dose patients can safely receive.
Relay is betting that other drugmakers haven't been able to target just FGFR2 because they're viewing it statically alongside other members of the FGFR family, which makes these proteins look nearly identical structurally. But Relay's Motion-Based Drug Design is rooted in the idea that the static structure of a protein only tells part of the story, and that insights into the way that protein moves in the body are essential for developing a molecule that can bind selectively to it.
When Relay executives give presentations about this approach, they show side-by-side, pared-down images of FGFR1 and FGFR2, which look identical. Then they show simplified clips of the simulations of the proteins in motion, and upon closer inspection, it is apparent that a portion of these proteins has a different motion.
"When we show this animation, people smile," Patel said, noting that while these simulations are a simplified version of the complex visuals that Relay researchers use to develop drugs, they easily convey what differentiates Relay's drug development process from that of other companies.
Now, the safety and maximum tolerated dose of Relay's highly selective RLY-4008 is being evaluated in a Phase I clinical trial that is enrolling patients with cholangiocarcinoma and solid tumors characterized by FGFR2 gene fusions, mutations, or amplifications. Patients are tested for these alterations using next-generation sequencing.
RLY-1971, which entered a Phase I clinical trial earlier this year for patients with solid tumors, is designed to bind to SHP2, a protein that promotes cancer cell survival and growth through the RAS pathway. While the patients enrolled in this study are not selected based on particular biomarkers, the company will likely home in on biomarker-defined patient populations as the drug moves into later-stage development. As a monotherapy, the drug may have activity against lung cancers driven by KRAS G12C mutations. In combination with tyrosine kinase inhibitors that target alterations in genes such as ALK or EGFR, RLY-1971 may prevent acquired resistance to those agents.
Right now, Relay is broadly enrolling patients in the Phase I trial in order to get a good read on RLY-1971's safety profile. However, exclusion criteria for the trial bar the enrollment of patients whose tumors have alterations that Relay already knows will not respond to RLY-1971, including BRAF V600E mutations, MEK mutations, or KRAS G12D, G12V, G13X, or Q61X mutations. Within the Phase I trial, Relay plans to further explore how RLY-1971 impacts patients' circulating tumor DNA, as well as track preliminary response rates to inform future, biomarker-based patient selection. The company expects to share a clinical update on the RL-1971 program in 2021.
Finally, the agent that Relay has in preclinical development, RLY-PI3K1047, is believed to be more selective than other agents targeting PI3Kα. Similar to the claims that Relay has made about its FGFR2 inhibitor RLY-4008, the company believes that RLY-PI3K1047, by being more selective, will be able to avoid the toxicities seen with other PI3Kα inhibitors that inhibit wild-type PI3Kα and other PI3K isoforms.
Next steps, long-term goals
As of now, Relay has not entered into partnerships with other drugmakers to clinically develop or commercialize its programs, but "all options are on the table," according to the company. Relay may seek out partnerships to develop RLY-1971 in combination with other agents, for instance.
And beyond the specific agents in its pipeline, Relay sees areas for potential collaborations down the line, possibly applying its Dynamo platform toward developing precision drugs in non-cancer settings, including immunology and neurology.
The expansion beyond precision oncology is already in the works, since the company launched two genetic disease programs earlier this year. For competitive reasons, the details of these programs, as well as three oncology programs not yet listed in its pipeline, remain under wraps for the time being.